Research On The Recognition Of Coal Gangue Based On Image Enhancement For Top Release Coal Mining

As one of the key technologies of integrated coal release mining,intelligent improvement of its technology is a very important part in the process of intelligent mine construction,and gangue identification technology is a necessary step in the intelligent improvement of coal release,so it is very important to study the application in the field of coal gangue identification of coal release.In this thesis,we mainly study the gangue recognition technology based on image enhancement technology,compare and analyze the target detection and segmentation algorithms that have practical application in engineering,and select the YOLOv5 s network model with fast training speed and comprehensive performance as the target detection algorithm for gangue recognition.The image quality enhancement algorithm and image data set enhancement algorithm are selected for the defects of coal gangue images,including low illumination,dust and fog,homogeneity and other problems of field collection images.It is also necessary to consider how to optimize the original network model for the characteristics of coal gangue recognition applications.In order to solve these problems,the following studies are conducted in this thesis:(1)By analyzing the features and differences on the images of coal blocks and gangue,we determine the features to be extracted when performing gangue recognition,including grayscale features,edge features and texture features.At the same time,analyze the existing three coal gangue recognition methods: morphological watershed algorithm,Faster-RCNN,and YOLOv5,according to the coal gangue recognition experimental results to analyze the advantages and disadvantages of three coal gangue recognition algorithms have,YOLOv5 s algorithm has the best performance,the accuracy rate,recall rate and detection time are 86.74%,88.97% and 0.14 s.(2)The solution of coal gangue image enhancement is proposed for the problems of low illumination and many noise points possessed by coal mine production working face images,by comparing the advantages and disadvantages of four coal gangue image quality enhancement methods,such as median filtering,bilateral filtering,dark channel defogging and multi-scale MSR,and verified by experiments,according to the comparison results of image quality evaluation indexes,the dark channel defogging method performs the best,and the information entropy,peak signal-to-noise ratio and The information entropy,peak signal-to-noise ratio and structural similarity reach 7.6249,48.0984 and 0.8354,respectively.(3)In view of the problems of small data volume and high data homogeneity of coal mine production working face,the solution of coal gangue image dataset enhancement was proposed,and the advantages and disadvantages of two coal gangue image dataset enhancement methods,single sample geometric transformation and multiple sample Mixup enhancement,were compared and analyzed,and Mixup enhancement was selected as the coal gangue image dataset enhancement method for further research in this thesis by combining the characteristics of coal mine production working face image dataset.Enhancement method.(4)For the optimization and improvement of YOLOv5 s network model,through the analysis of the main basis of gangue identification,the characteristics of the gangue image dataset and the characteristics of YOLOv5 s network model itself,the CBAM attention mechanism is selected to optimize the original network model.And using the laboratory high illumination gangue image,simulated low illumination gangue collection image,field collection of top coal gangue image three kinds of images on the improved YOLOv5 s network model for experimental verification,the results show that the performance of the YOLOv5 s network model with the addition of attention mechanism compared to the original network model has been improved,m AP@.5:95,Val/box＿loss,Val/cls＿loss improved by 3.9%,3.7%,and 2.7%,respectively.